Sobol’ sensitivity analysis of a complex environmental model

Complex environmental models are controlled by a high number of parameters. Accurately estimating the values of all these parameters is almost impossible. Sensitivity analysis (SA) results enable the selection of the parameters to include in a calibration procedure, but can also assist in the identification of the model processes. Additionally, a sensitivity analysis can yield crucial information on the use and meaning of the model parameters.This paper presents a Sobol’ sensitivity analysis for flow simulations by a SWAT model of the river Kleine Nete, with the objective to assess the first order, second order and total sensitivity effects. Confidence intervals for the resulting sensitivity indices are inferred by applying bootstrapping. The results indicate that the curve number value (CN2) is the most important parameter of the model and that no more than 9 parameters (out of 26) are needed to have an adequate representation of the model variability. The convergence of the parameter ranking for total sensitivity effects is relatively fast, which is promising for factor fixing purposes. It is also shown that the Sobol’ sensitivity analysis enhances the understanding of the model, by e.g. pointing out 3 significant pairwise interactions.In general, it can be concluded that the Sobol’ sensitivity analysis can be successfully applied for factor fixing and factor prioritization with respect to the input parameters of a SWAT model, even with a limited number of model evaluations. The analysis also supports the identification of model processes, parameter values and parameter interaction effects.

► First order, second order and total Sobol’ sensitivity indices are assessed.
► Confidence intervals and a repeatability test are used to evaluate the results.
► 9 SWAT parameters are necessary to adequately represent the model variability.
► Sobol’ can be applied for factor fixing with a limited number of SWAT simulations.
► The Sobol’ SA supports the identification of model processes and parameters.